The present invention relates to data storage cache management and more specifically to managing a cache in a hybrid pool containing storage volumes having different latencies.
Low latency storage arrays provide sub millisecond response times but are expensive. Virtualization systems can intelligently place hot data on the expensive low latency storage array subsystems and cold data on cheaper, higher latency storage array subsystems. Many virtualization systems provide a redundant cache to hide the latency of the slower subsystem, but with software defined storage devices using commodity hardware, the latency added due to replication of the data between the sides of the redundant cache is greater than the latency of writing directly to the low latency storage array subsystem.
Prior art solutions include the use of non-commodity hardware with dedicated low latency links between the redundant caches, or more expensive network technologies such as Infiniband, promoted by the Infiniband Trade Association. Another prior art solution is using only low latency storage and disabling cache in the virtualization system. All of these solutions are expensive.
Further prior art solutions are to manually configure volumes to place hot data on the low latency storage and to disable cache for those specific segments. These are complicated, require constant management and are liable to user error.